[BigMackCam]

EDIT: Thread moved from members' articles to Pentax DSLR Discussion per Adam's advice, for greater visibility...

Despite having been released in 2006, the Pentax K10D and its cousin, the Samsung GX-10, remain popular cameras even now in 2019. A major reason for this popularity is the claim by fans that the CCD sensor produces more pleasing colour and tones when compared to later CMOS sensor Pentax models.

I decided to perform a quick test to demonstrate the difference in colour and tone rendering between my Samsung GX-10 and Pentax K-3, as I felt it may be of interest to other members.

This was not a scientifically-controlled test, though I did attempt to make the conditions close to identical for both cameras.

I placed my X-Rite ColorChecker chart on my dining room table. Facing towards the window. It was a dull and very cloudy afternoon (the sky was completely greyish-white). I also had the overhead room light turned on, but this was not shining directly onto the target, being some way behind and above it.

Both GX-10 and K-3 were set to their sRGB colour space, "Natural" profiles, with DNG selected as the file format, ISO set to 100, and AWB white balance.

Using the same SMC Pentax-DA 35mm f/2.4 lens, I took almost identical shots of the ColorChecker chart with each camera, differing only slightly in angle and distance since these were taken hand-held.

I imported the resulting files into Darktable 2.4.4. For each photo, I:

- disabled all default processing options (base curve etc.) that might affect colour, shadows and highlights, black and white points, contrast, tone curve and so on
- set input color profile to "embedded matrix"
- set demosaic to AMaZE
- cropped and rotated to select the same area of the chart
- set white balance by spot selecting the same mid grey tone on the chart

Both shots were slightly under-exposed, so I adjusted them such that the luminosity of the mid-grey tones matched. (using the color picker tool for measurement). I then re-performed white balance adjustment, just in case.

Both shots were exported with output color profile set to sRGB (web-safe).

See below the results of the test, displayed as an animated GIF... please excuse the compression artefacts :

The difference in colour and tone is clear, with the GX-10 photo looking brighter, warmer and richer. Indeed, the GX-10 photo requires little further processing. The K-3 photo, by comparison, looks flatter... lacking in saturation; almost pastel-like. It requires more post-processing to achieve a pleasing result.

Some believe that the GX-10 / K10D sensor has a yellow cast, but it's not that simple. White balance has been normalised for both shots, and we can see that the white-grey-black squares match very closely, with no obvious colour tint. It's true that most of the colours have a warmer hue, but saturation and luminosity are also quite different. For example, the GX-10's reds are warmer, brighter and more saturated, while the blues are more saturated yet darker. Clearly there's a lot more to the differences than a simple colour tint.

Is the CCD sensor responsible for this difference in colour and tone? In the past, I've said that I prefer the "look" of photos from cameras with CCD sensors, but it's difficult to say conclusively whether the sensor alone should be credited for the rendering. I believe the sensor is partly responsible, but feel the camera's colour matrix is also a significant factor.

Whatever the reasons, I continue to prefer the colours and tones from my GX-10 raw files. I'm in no way unhappy with the K-3's files, though. Quite the contrary... They respond well to post-processing, and it's possible to achieve similarly appealing results. It just takes a little more work

As a final note, I'm unaware if the color matrices for the GX-10 and and K10D are the same, or tweaked slightly by each manufacturer. I don't own a K10D, so can't compare. However, it seems reasonable to assume they are the same, or at least quite similar.

I hope this is of interest to some of you

---

Last edited by BigMackCam; 02-07-2019 at 01:35 AM.

[Igor123]

Great experiment and interesting result! (I had only heard the discussion about that but never looked more at it)

To my understanding, wouldn't the color matrix "just" be the cameras own way to tell the software using it how the camera/maker thinks the raw values "should" be mapped (to a certain color space) to arrive at the intended (by the maker) colors? (So each raw converter could have it's own default profile for each camera model affecting this result?)

If so I guess the not-possible-to-answer-objectively-yet-is-the-$1000-question is =) : Which one looks more (on your screen) like what you see (with your eyes)?(parentheses included for emphasizing that those will be variables that may differ and alter the result others do the same experiment...)

And if using the default profiles in DT, is the difference as pronounced, or have they "equalized" the extremes?

Edit: And if they are not as different with default profiles, are they "in the middle" or do the DT profiles "prefer" one of the looks over the other?

[The Squirrel Mafia]

There's something magical about that old 10MP Sony sensor. I have also compared similar shots from my old Sony a200 against the Pentax K-50. Similar thing. At ISO 100, the 10MP sensor is amazing. The colors are on a different level of amazing. I can't replicate it with the K-50.

The area where my K-50 completely destroys the a200 is at higher ISO. The max I rarely ever used on the a200 was ISO 800. I can go up to ISO 12800 on the K-50 quite easily.

[Dartmoor Dave]

That definitely matches my own experience, and I can't really add anything other than to say that I agree. CCD seems to produce raw files that look good straight off the camera, while CMOS needs much more processing.

I think the difference in the sensor technologies is part of it, but I think it might also have to do with different secret sauces being cooked into the raw files. My theory is that the Pentax CCD cameras were made at a time when manufacturers were still trying to lure serious photographers away from film, so the cameras were designed to give the most slide film-like look possible. While modern CMOS cameras are designed around a more modern "digital" look, with resolution and noise performance seen as much more important than giving a rich colour rendering straight off the camera. You can get great results with both technologies, but CMOS seems to need more work.

As a side-note, I've compared my GX-10 with a friend's K10D using the same lenses in exactly the same light, and we couldn't detect any difference in the raw files. So I'm convinced that the only differences in the cameras are cosmetic.

[TonyW]

Interesting comparison tests. My belief is that the major contributor here is not necessarily related to CCD vs CMOS, but one of the raw rendering engine in your editor of choice.

While you can set white balance based on white and grey patches of a test object it does not guarantee a pleasing or necessarily accurate response to other colours. AFAIK there is no standard way that raw data must be converted per camera. So a conversion could conceivably give absolute neutrals with Hue and Sat twists and shifts for other colours. If you look at a raw file in its natural state before demosaic, WB, TRC etc it will probably not be too surprising that different rendering between raw engines occur - similar to the ‘look’ applied to manufacturers generated JPEG picture controls

[paulh]

Mike, thanks for taking the time to document your little experiment - quite interesting!

[Dartmoor Dave]

Here's a couple of handheld comparisons shot back in 2016. In both cases the top camera is the Samsung GX-10 version of the K10D, and the bottom camera is the K-S1. The lens is my personally most used one: the Super Takumar 20mm/4.5.

This first example is raw files converted to jpeg in Camera Raw using the embedded profiles. White balance is taken from a white card for each camera.



In this second example the raw files are converted in Camera Raw using my own custom profiles, created in Adobe Profile Editor with colour correction and linear tone curves.



It seems to me that the differences between the sensor technologies are much less significant using the custom profiles, but then of course the whole point of using Adobe Profile Editor is to minimise the effect of the manufacturer's own profiles. However, the difference is still there. It's not so easy to see just from a couple of test shots, but in day-to-day use there is absolutely no doubt in my mind that the CCD sensor gives richer colours and a more "luminous" effect of shadow and light. And the difference remains, no matter how much I play around with profiles to get the look I want from CMOS. That's why I still shoot with a 10MP CCD unless high ISO is a necessity.

[BigMackCam]

To my understanding, wouldn't the color matrix "just" be the cameras own way to tell the software using it how the camera/maker thinks the raw values "should" be mapped (to a certain color space) to arrive at the intended (by the maker) colors?

Yes, that's my understanding too. And one problem, therein, is that different raw processors render colour and tone information somewhat differently - and the camera manufacturer's matrix can't account for that. In that sense, I see the manufacturer's matrix as a starting point that at least gives the internal JPEG engine and any external raw conversion software a starting point, based on the manufacturer's interpretation of colour.

(So each raw converter could have it's own default profile for each camera model affecting this result?)

...
And if using the default profiles in DT, is the difference as pronounced, or have they "equalized" the extremes?

Each raw processor, or the libraries on which the raw processor is based (such as dcraw, for example), will often have custom profiles available for various cameras, and will allow you to select one of these instead of the camera's embedded matrix. In many (perhaps most) cases, I believe these custom profiles are better tuned to render well in the relevant software.

There's something magical about that old 10MP Sony sensor. I have also compared similar shots from my old Sony a200 against the Pentax K-50. Similar thing. At ISO 100, the 10MP sensor is amazing. The colors are on a different level of amazing. I can't replicate it with the K-50.

The area where my K-50 completely destroys the a200 is at higher ISO. The max I rarely ever used on the a200 was ISO 800. I can go up to ISO 12800 on the K-50 quite easily.

I'm very happy shooting the GX-10 up to ISO 400. I'll push it to 800 if absolutely necessary, but I prefer not to. At ISO 400 or below, though, it's great. At ISO 100, it's wonderful. For higher ISO work, I'd pick up my K-3 every time...

As a side-note, I've compared my GX-10 with a friend's K10D using the same lenses in exactly the same light, and we couldn't detect any difference in the raw files. So I'm convinced that the only differences in the cameras are cosmetic.

Thanks, Dave, that's good to know. I hoped and suspected it might be the case.

My belief is that the major contributor here is not necessarily related to CCD vs CMOS, but one of the raw rendering engine in your editor of choice.

While you can set white balance based on white and grey patches of a test object it does not guarantee a pleasing or necessarily accurate response to other colours. AFAIK there is no standard way that raw data must be converted per camera. So a conversion could conceivably give absolute neutrals with Hue and Sat twists and shifts for other colours. If you look at a raw file in its natural state before demosaic, WB, TRC etc it will probably not be too surprising that different rendering between raw engines occur - similar to the ‘look’ applied to manufacturers generated JPEG picture controls

Thanks for that, Tony. I can confirm that I get very similar results in RawTherapee, and I believe - though I'm not 100% certain - that darktable and RT use different raw conversion libraries. So I'm reasonably confident that, whilst the raw converter may have a role to play, the differences are already there in the data.

Furthermore, although I chose to normalise white balance in post-processing, the two shots weren't that far apart to begin with. Even using the cameras' white balance settings recorded in the raw files, you can see the difference in H, S & L of each colour. The hues are very slightly different because of the relatively minor white balance difference, but not significantly. And, since, white balance is independent of the recorded raw data, had I set custom white balance in-camera using the same grey square on the chart (an excellent, reliable target in my experience), I believe the results out of both cameras would closely match the results presented here.

Mike, thanks for taking the time to document your little experiment - quite interesting!

You're welcome, Paul

It seems to me that the differences between the sensor technologies are much less significant using the custom profiles, but then of course the whole point of using Adobe Profile Editor is to minimise the effect of the manufacturer's own profiles. However, the difference is still there. It's not so easy to see just from a couple of test shots, but in day-to-day use there is absolutely no doubt in my mind that the CCD sensor gives richer colours and a more "luminous" effect of shadow and light. And the difference remains, no matter how much I play around with profiles to get the look I want from CMOS. That's why I still shoot with a 10MP CCD unless high ISO is a necessity.

I agree, Dave. The custom profiles provided with each raw processor seem to bring the results closer together. Sadly, there is no custom profile in darktable for the GX-10, but there is for the K-3 - and using this, the results are somewhat closer to the GX-10. Perhaps if there was such a profile for the GX-10, this would render in a similar fashion. In any case, see below a GIF illustration which now includes the K-3 using embedded matrix, K-3 using DT's standard matrix, and GX-10 using embedded:



Though somewhat "quick and dirty", I believe this test serves a useful, if narrow, purpose... to show that the raw output from these two cameras, when processed through the same raw converter using the manufacturers' embedded matrices, is rendered differently in terms of hue, saturation and luminosity. Whether that's due to the sensor, the manufacturers' matrices, or both, I really can't say for sure. I suspect it's both, since I find it difficult to believe that two sensors employing different technology to record light would do so in an entirely consistent fashion. Whatever, the end results are very different, and I prefer what I see from the old GX-10

[Igor123]

Y
...

Lots of insightful answers, and you said you preferred the CCD output, but you avoided my "$1000" question =), since you have seen the chart with your own eyes; which output do you think resemble the real thing? Is the CCD closer to the truth, or does it over-saturate the colors compared to what you see?

Thanks!

[BigMackCam]

Lots of insightful answers, and you said you preferred the CCD output, but you avoided my "$1000" question =), since you have seen the chart with your own eyes; which output do you think resemble the real thing? Is the CCD closer to the truth, or does it over-saturate the colors compared to what you see?

My apologies, Igor

Based on the embedded matrix for each camera, and processing in Darktable (also in RawTherapee), I don't believe either gives a completely "realistic" reproduction of what we see. The GX-10 / K10D output is generally warmer and richer than reality... I might call it "optimistic" ... yet the reds are more accurate, and blue skies are far more accurate. The K-3 output is cooler and muted, with darker, magenta-like reds and poorly saturated blue skies... "pessimistic", perhaps.

To get the most accurate reproduction of colour and tone, I believe we would need to produce dual-illuminant profiles using controlled lighting conditions for each camera and raw converter. That would, to some extent, normalise output from both cameras based on the colour test chart used in profiling. Yet I still think there would be differences, since I don't believe the sensors are behaving in precisely the same way - even between two brands of CMOS sensor, let alone an old CCD sensor and a modern CMOS variety.

With the GX-10 / K10D, I believe the output is attractive and needs little colour or tonal work, though it's not completely accurate. But I'm not sure that I'd want it to be. It's rather like a really nice 35mm film in that respect. With the K-3, the output needs work whether you want it to look accurate or like a nice 35mm film. Does that make sense?

[Igor123]

My apologies, Igor

...
Does that make sense?

I think it does. I guess we need to start requesting factory-calibrated sensors or calibration tools like for our screens!

Joking aside, or not, perhaps that could be a differentiator between brands if a "ISO standard" or something like that for sensor color accuracy (including the profile) existed and cameras or sensors could be rated according to how accurate they are..like say audio amplifiers that have snr, thd, imd noise metrics etc. Then there will always be, like in audio, proponents of more subjective characteristics so not all will agree that the most "accurate" sensor is the best, but that's another discussion...

But either way I'm at a stage where I don't have much time for "quality" photography but end up doing mostly "documentation" of family events/vacations etc, so I don't spend the time on the color work that I would perhaps want/need to be close to this being a top priority. In few years perhaps =). Instead just reading and commenting about it and dreaming that one day I will make my own color profile...=)

[BigMackCam]

Just a thought... Should anyone wish to carry out a similar test with their K10D and later CMOS camera using Pentax Digital Camera Utility, I'd be very interested and keen to see the results

[swanlefitte]

It's very interesting to see the overlay. I was most intrigued looking at the 2 primary color schemes. Rgb had major blue, some red, and hardly any green change. The other way yellow had a little, cyan a little, but magenta had quite a bit of change.

[Dartmoor Dave]

I agree, Dave. The custom profiles provided with each raw processor seem to bring the results closer together. Sadly, there is no custom profile in darktable for the GX-10, but there is for the K-3 - and using this, the results are somewhat closer to the GX-10. Perhaps if there was such a profile for the GX-10, this would render in a similar fashion. In any case, see below a GIF illustration which now includes the K-3 using embedded matrix, K-3 using DT's standard matrix, and GX-10 using embedded:

Just to clarify: the custom profiles that I used in the bottom comparison where the results look very similar were not provided with the raw processor. They are custom profiles that I created for the two actual camera bodies that I own, by shooting colorcheck cards at 6500K then using Adobe Profile Editor's "chart" function to generate dual illuminant profiles. The purpose for doing this is exactly that it removes the impact of the embedded profile (or matrix as you call it) from the results. So that any difference that you're left with is down to the sensor technology itself rather than the way the raw data is interpreted by the processor. And remember that these are camera profiles that adjust the interpretation of the raw data from the sensor; they are not just processing presets.

My overall point is that most of the obvious differences that we see between CCD and CMOS are caused by the manufacturer's embedded profiles and the way those profiles are interpreted by raw processors. However, once you remove as much of the impact of the embedded profiles as possible, there is still a difference that can only be ascribed to the sensor hardware itself. The actual hardware difference is a lot more subtle than the obvious differences caused by the profiles, but it's still there. And it's significant enough that I still shoot with CCD.

The test card examples that you've posted in this thread look to me like they would work in Adobe Profile Editor to create your own custom profiles, so if Darktable is able to use .dcp profiles in its raw convertor it might be worth giving it a try. Or send me the raw files that you used and I'll happily have a go at generating the profiles myself.

[K1N8]

Would any build in image processing setting of the K3 come close to the results from the CCD?